Canva launches Craft, its first image-generation model

A model trained on design work

Canva trained Craft on designers’ work from its template library, including:

Images generated with Craft can be turned into layered designs using Canva Magic, the company’s tool for converting images into designs. Last year, Canva released its own model for recognizing layouts and layers.

Researchers from Leonardo.ai also helped develop Craft. Canva acquired the media-generation platform in 2024.

Control over cost and speed

Canva previously used several different image-generation models on its platform. A model of its own could reduce the company’s API costs and give it more control over generation quality and speed.

The company is also working on two less expensive versions, Craftfast and Craft-ultrafast, which it plans to release soon.

The business question

Canva had expected $4 billion in revenue in 2024, but lowered its growth forecast for 2026. External analysts estimate that its valuation has fallen from $42 billion to $35 billion.

I think Craft’s importance is less about adding another image generator than about bringing a core capability in-house: Canva can shape the cost, speed and design workflow instead of relying on a mix of outside models. But the announcement gives no benchmarks for quality or generation speed, and no figures for expected savings. Until those appear, the business case is a direction, not a demonstrated result.